• Everyday AI
  • Posts
  • Ep 785: What’s new in Gemini 3.5 Flash, Google Omni and Antigravity 2.0: Hands On With the latest from Google I/O

Ep 785: What’s new in Gemini 3.5 Flash, Google Omni and Antigravity 2.0: Hands On With the latest from Google I/O

Robinhood lets agents control your investing, Microsoft's big splash in AI images, Google's next enterprise AI play, NotebookLM is finally getting auto-syncing Google Drive files and more

Author

Everyday AI
May 27, 2026

Outsmart The Future

Today in Everyday AI
8 minute read

🎙 Daily Podcast Episode: Google just dropped a flood of AI updates at I/O, and beneath the flashy demos are some big shifts in pricing, workflows, and what “good enough” AI actually looks like for businesses. Give today’s show a watch/read/listen.

🕵️‍♂️ Fresh Finds: Claude can now scan and fix risky code before it hits production, TikTok and Universal Music Group are teaming up against unauthorized AI music, and NotebookLM is finally getting auto-syncing Google Drive files, and more.Read on for Fresh Finds.

🗞 Byte Sized Daily AI News: Robinhood lets agents control your investing, Microsoft's big splash in AI images, Google's next enterprise AI play and more. Read on for Byte Sized News.

💪 Leverage AI: Google I/O didn’t just launch new AI tools. It made AI strategy a lot messier for businesses trying to figure out what’s actually useful, affordable, and ready for real work. Keep reading for that!

↩️ Don’t miss out: Miss our last newsletter? We covered: Qualcomm just landed a major AI chip deal with ByteDance, China is tightening restrictions on top AI talent, and world leaders are increasingly debating how AI should shape jobs, power, and society, and more.Check it here!

Ep 785: What’s new in Gemini 3.5 Flash, Google Omni and Antigravity 2.0: Hands On With the latest from Google I/O


You’ll need a map, compass and legend to understand all the new AI Google announced at its I/O conference last week.

(They literally wrote a blog post called, "100 things we announced at I/O 2026” and most of them were AI based.)

Luckily for you, we spend hours each day going through the latest in AI to cut the fluff from the real. So on today’s ‘AI Working Wednesdays’ series, we break down 3 of Google’s biggest AI updates you can use today: Google Omni, Gemini 3.5 Flash and Antigravity 2.0.

What’s new and how do they work? We’ll show you the ins and outs live.

Also on the pod today:

• Gemini 3.5 Flash now pricier 💸 
• Anti Gravity 2.0 desktop debut 🖥️
• Omni Flash: Not just video 🎥 

Listen on our site:

Click to listen

Subscribe and listen on your favorite podcast platform

Listen on:

Here’s our favorite AI finds from across the web:

New AI Tool Spotlight –  Bluedot captures, transcribes, and summarises every meeting, online or in-person, Powabase is the all-in-one development platform for AI apps, CometChat Gives your AI coding agents the power to integrate in-app chat and calling without the guesswork.

Claude Security — Claude can now quietly check its own code for risky stuff like injection, unsafe deserialization, and sketchy DOM calls, then fix issues before they hit a PR.

OpenAI Taxes — OpenAI says Tax AI got better over time by turning real practitioner corrections into structured evals and Codex tasks, not by waiting on engineers to debug every miss.

Tiktok and Universal Music Group — TikTok and Universal Music Group just renewed their deal, and this time they’re also teaming up to remove unauthorized AI-made music.

Figure and Catalyst — Figure just landed a commercial deal with Catalyst Brands, the retailer behind JCPenney, Aéropostale, and Brooks Brothers.

ElevenLabs Music v2 — ElevenLabs just launched Music v2, with stronger vocals, better instrumentation, and smoother genre shifts across a wider range of styles.

OpenAI Sunsetting Models — OpenAI is sunsetting GPT-5.2 and GPT-5.3-Codex in Codex for ChatGPT users on June 2, with GPT-5.5 taking over for free plans.

Claude Voice Update — Anthropic’s testing a bigger Claude voice update, with a new UI, push-to-talk, and a beta language menu that already lists several non-English options.

NotebookLM — NotebookLM is getting a big upgrade, Google Drive files will now auto-sync instead of needing constant re-uploads.

Baseten Valuation — Baseten is reportedly in talks to raise $1 billion at an $11 billion valuation.

Alook Open Source — Alook just went open-source with a setup that lets one person run a whole AI team through a simple org chart.

Claude Containment — Anthropic’s takeaway is pretty simple: agent safety is less about trusting the model and more about boxing it in with tight sandboxes, VM limits, and network controls.

DuckDuckGo and Google Search — Google’s AI-heavy Search overhaul is pushing some users toward DuckDuckGo, which says installs jumped after the backlash.

1. NVIDIA plans Taiwan HQ with $150B push 💸

NVIDIA said Wednesday it will build a Taiwan headquarters, with CEO Jensen Huang calling the island the “epicentre” of the AI boom and saying the project should be operational by 2030.

The chip giant says it plans to invest about $150 billion a year in Taiwan, deepen ties with key partners like TSMC, Foxconn, Wistron, and Quanta, and hire about 4,000 people at the new site.

2. Zuckerberg-backed AI targets drug discovery 💊

A new AI “world model” from a Zuckerberg-backed philanthropic venture is aiming to speed up drug discovery, putting new attention on how fast-moving AI is being applied to medicine.

The effort matters because it tries to help researchers predict and test complex biological outcomes more efficiently, which could shorten early-stage development work. According to MSN, this is part of the growing race to use AI for scientific research, not just chatbots and content tools.

3. Leak: Google widens Gemini for Business with shared projects and automated agents 📲

Google is reportedly pushing Gemini for Business closer to its Enterprise version with new team-focused tools now moving from internal testing toward wider rollout.

The biggest update is Projects, which gives users shared workspaces with their own chats, files, colors, and system instructions, making Gemini feel more like a coordinated team hub than a single chat app.

4. MAI-Image-2.5 debuts at No. 3 on Arena 🥊

Microsoft’s MAI Superintelligence Team has just unveiled MAI-Image-2.5, and the timing matters because it lands already ranked third on the Arena text-to-image leaderboard.

The company says the new model is its strongest yet, with better instruction-following, sharper text rendering, and more coherent images across styles.

5. Biohub Unveils 1B-Protein AI Atlas 🤑

A newly released preprint from the Chan Zuckerberg Biohub says its open-source ESMFold2 system has generated an atlas of 1.1 billion predicted protein structures and 6.8 billion protein sequences, making the database far larger than AlphaFold’s.

The timing matters because this lands in a fast-moving race to map biology with AI, and Biohub is claiming its model can do better than rival systems at predicting how proteins fold and interact, including antibodies. In plain terms, that means scientists may now have a much bigger reference map for finding useful biology hidden in poorly understood proteins.

6. GPT-5.5 Crushes in new SWE Benchmark 💻

Its DeepSWE test found OpenAI’s GPT-5.5 far ahead of the pack, while also claiming that SWE-Bench Pro often grades solutions incorrectly and can be gamed by models that inspect hidden repository history. In plain terms, Datacurve says the industry may have been judging coding agents with a scoreboard that is both too forgiving and too unreliable.

7. Robinhood lets AI trade and spend for users 📈

Robinhood just unveiled a new push into autonomous finance, announcing tools that let AI agents trade stocks and make purchases on behalf of retail customers.

The company said the system, called Agentic Trading and an Agentic Credit Card, can follow user instructions with limited human involvement, while keeping some guardrails like separate accounts, alerts, spending limits, and the ability to cut off access fast.

Google just made AI strategy more expensive.

Not because every new tool costs more.

Because after Google I/O, the real tax is figuring out what’s actually useful, what’s available, what quietly changed under the hood, and what looks impressive until your team tries to run work through it.

That’s where leaders get smoked. A 100-announcement AI drop sounds like progress, but for executives it creates a new operating problem: your team can now move faster in the wrong direction, with shinier tools, fuzzier limits, and a model name that might not mean what it used to mean.

Fun.

On today’s Everyday AI, we put Gemini 3.5 Flash, Gemini Omni Flash, and Google Antigravity 2.0 under the only microscope that matters: should your business actually use this, where does it unlock leverage, and where is Google asking you to bring your own common sense?

Let’s save you a few meetings.

1. Stop trusting model labels 🔥

Gemini 3.5 Flash is good enough to cause bad decisions.

That’s the trap.

It’s fast, capable, strong on reasoning tests, and powerful enough to make many teams assume it’s an easy upgrade from older Flash models. The problem is the word “Flash” used to carry a pretty obvious meaning: fast and cheap.

Now? Fast, yes. Cheap, nah.

For leaders, this is bigger than Google naming weirdness. If your AI roadmap depends on model swaps, agentic workflows, API usage, or internal tools running quietly in the background, then “better model” is only half the math.

The other half is cost per completed task. Gemini 3.5 Flash can be 10 to 20x more expensive than previous Flash versions on the API side, and if your team treats it like the old cheap workhorse, your AI budget won’t explode loudly.

It’ll leak.

Try This

Before anyone swaps Gemini 3.5 Flash into production, run a routing audit across three real workflows: support, research, and coding.

Compare it against your current model on task completion, token use, latency, and whether the output actually reduces human cleanup. Use it where speed and capability create obvious value. Keep it out of low-stakes jobs where “pretty good but pricier” becomes the most boring budget mistake imaginable.

2. Use Omni beyond video ⚡

Most teams will look at Gemini Omni Flash and ask the wrong question.

They’ll ask whether the video looks better than another video model.

Cool. Also too small.

Omni matters because it takes text, image, audio, and video inputs, then creates editable video through conversation. That means the strategic unlock isn’t just “make us a clip.” It’s turning messy business communication into something teams can reshape without waiting on the usual approval swamp.

Think training content. Product explainers. Sales enablement. Internal updates. Customer education.

The transcript’s bigger signal is that Omni should be judged as a workflow layer, not a novelty generator. Yes, the clips are short. Yes, video-to-video can still wobble. But when a model can understand multiple input types and let users edit the output conversationally, the advantage goes to companies that redesign content operations around iteration speed.

Not vibes. Throughput.

Try This

Pick one recurring asset your team makes every month and map the entire path from rough idea to published version.

Count every handoff, rewrite, export, review, and “can you make this small change?” moment. Then test whether Omni can collapse the middle of that process. The goal isn’t to replace your creative team; it’s to stop making talented humans babysit tiny revisions that AI should eat for breakfast.

3. Make agents prove the finish 🚀

Google Antigravity 2.0 is the part leaders should watch without falling in love yet.

It has the right ingredients: desktop app, local folder access, scheduled tasks, multiple models, parallel agents, and a pitch that screams future of work.

Then reality shows up wearing sweatpants.

The app can produce useful front ends and move quickly, but the important business lesson comes from the gap between “it built something” and “it finished the job.” An agent that creates a slick interface with broken or misdirected links didn’t deliver a business outcome. It delivered a demo with homework attached.

That’s where execs need to get ruthless. Agentic coding tools like Codex, Claude Code, Cursor, and now Antigravity should be evaluated on completed workflows, validated outputs, and clean handoffs, not how magical the first screen looks.

Pretty is cheap now.

Working is the moat.

Try This

Create an agent acceptance test before your team standardizes on any coding agent.

Require five checks: correct source data, working links, tested actions, clear launch instructions, and proof that the output solves the original user problem. Let teams experiment with Antigravity, especially if they’re already deep in the Google ecosystem, but don’t confuse “agent made a thing” with “business got value.”

That gap is where AI strategy either gets real or gets expensive.

Reply

or to participate.